B1992
Title: Risk control for online learning models
Authors: Yaniv Romano - Technion---Israel Institute of Technology (Israel) [presenting]
Abstract: Modern machine learning algorithms have achieved remarkable performance in a myriad of applications, and are increasingly used to make impactful decisions in the hiring process, criminal sentencing, and healthcare diagnostics. The use of data-driven algorithms in high-stakes applications is exciting yet alarming: these methods are extremely complex, often brittle, and notoriously hard to analyze or interpret. Naturally, concerns have been raised about the reliability of the output of such machines. The focus is on making reliable predictions in an online setting, in which the underlying data distribution can drastically---and even adversarially---shift over time. We will introduce statistical tools that can be wrapped around any online ``black-box'' machine learning model to provide valid and informative uncertainty estimates.